Molecular Dissection of Somatic Skeletal Disease in Neurofibromatosis Type 1

Nandina Paria; Aysha Khalid; Bo Shen; Ben Lemoine; Jinyan Chan; Yared H Kidane; Ila Oxendine; Reuel Cornelia; Carol A Wise; Jonathan J Rios

doi:10.1002/jbmr.4755

. 2022 Dec 13;38(2):288–299. doi: 10.1002/jbmr.4755

Molecular Dissection of Somatic Skeletal Disease in Neurofibromatosis Type 1

Nandina Paria ^1,^†, Aysha Khalid ^1,^†, Bo Shen ², Ben Lemoine ³, Jinyan Chan ^3,⁴, Yared H Kidane ¹, Ila Oxendine ¹, Reuel Cornelia ¹, Carol A Wise ^1,^5,^6,⁷, Jonathan J Rios ^1,^5,^6,^7,^✉

PMCID: PMC9898201 NIHMSID: NIHMS1854411 PMID: 36459048

ABSTRACT

Neurofibromatosis type 1 (NF1) is a tumor predisposition syndrome caused by heterozygous NF1 gene mutations. Patients with NF1 present with pleiotropic somatic secondary manifestations, including development of bone pseudarthrosis after fracture. Somatic NF1 gene mutations were reproducibly identified in patient‐derived pseudarthrosis specimens, suggesting a local mosaic cell population including somatic pathologic cells. The somatic cellular pathogenesis of NF1 pseudarthroses remains unclear, though defects in osteogenesis have been posited. Here, we applied time‐series single‐cell RNA‐sequencing (scRNA‐seq) to patient‐matched control and pseudarthrosis‐derived primary bone stromal cells (BSCs). We show that osteogenic specification to an osteoblast progenitor cell population was evident for control bone‐derived cells and haploinsufficient pseudarthrosis‐derived cells. Similar results were observed for somatic patient fracture‐derived NF1 ^−/− cells; however, expression of genetic pathways associated with skeletal mineralization were significantly reduced in NF1 ^−/− cells compared with fracture‐derived NF1 ^+/− cells. In mice, we show that Nf1 expressed in bone marrow osteoprogenitors is required for the maintenance of the adult skeleton. Results from our study implicate impaired Clec11a‐Itga11‐Wnt signaling in the pathogenesis of NF1‐associated skeletal disease. © 2022 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).

Keywords: BONE MODELING AND REMODELING, CELLS OF BONE, INJURY/FRACTURE HEALING, MOLECULAR PATHWAYS – REMODELING, ORTHOPEDICS, OSTEOBLASTS

Introduction

Proper skeletal development and fracture healing require coordinated regulation of myriad molecular processes orchestrating the differentiation of diverse cell types, such as bone‐resorbing osteoclasts and bone‐forming osteoblasts. In long bones, endochondral ossification proceeds through osteochondral differentiation of bone‐derived stromal cells (BSCs), including from the periosteal or bone marrow niches, to osteoblasts. Time‐series bulk‐cell expression analysis (RNA‐seq) of human telomerase reverse transcriptase–immortalized bone marrow‐derived stromal cells (hTERT‐MSCs) have shown that osteogenic differentiation is associated with temporal activation of pro‐osteogenic molecular pathways and repression of pro‐adipogenic metabolic pathways.⁽ ¹ ⁾ Consistent with this, single‐cell RNA‐seq (scRNA‐seq) of mouse CD45⁻Ter119⁻CD31⁻ bone marrow cells identified divergent adipogenic and osteochondral cellular trajectories and showed that intermediate osteoblast progenitor cells (OPCs) were associated with activation of genes required for formation of extracellular matrix (ECM).⁽ ² ⁾ These and other studies highlight the dynamic regulation of osteogenic differentiation and how disruption of this process leads to fracture healing defects in mice.⁽ ³ , ⁴ , ⁵ ⁾

Neurofibromatosis type 1 (NF1, MIM:162200) is a dominant tumor‐predisposition syndrome.⁽ ⁶ ⁾ Patients with NF1 harbor heterozygous germline mutations in the NF1 tumor‐suppressor gene, encoding the RAS GTPase neurofibromin, and somatic NF1 gene mutations (ie, loss‐of‐heterozygosity) are associated with pleiotropic clinical manifestations through activation of the RAS–MEK–ERK signaling pathway.⁽ ⁶ ⁾ Long bone dysplasia is rare in children with NF1 but carries a high fracture risk, often resulting in pseudarthrosis.⁽ ⁷ ⁾ Multiple studies have demonstrated the presence of somatic NF1 gene mutations in samples collected from patient fracture pseudarthroses, including cells cultured from biopsies adjacent and distal to the fracture site.⁽ ⁸ , ⁹ , ¹⁰ , ¹¹ ⁾ Somatic mutations have not been detected in cells cultured from unaffected bones, such as the iliac crest, and osteotomies at unaffected sites (ie, proximal tibia) of the involved bone heal without forming pseudarthroses,⁽ ⁹ , ¹² ⁾ suggesting somatic lesions are localized to a region of the involved bone. The molecular pathogenesis of NF1‐associated pseudarthrosis remains unclear, though prior bulk‐cell RNA‐seq implicated activation of the RAS–MEK–ERK and other cancer‐associated pathways.⁽ ⁹ ⁾ Interpretation of these results is limited, however, due in part to the mosaic nature of cultured patient fracture‐derived BSCs. To address this, we previously reported scRNA‐seq of clonally isolated fracture‐derived NF1 ^−/− primary cells, which confirmed changes in gene expression previously identified from bulk RNA‐seq.⁽ ¹³ ⁾ Despite these efforts, the impact that somatic loss of NF1 has on osteogenic differentiation of patient fracture‐derived BSCs remains unclear.

scRNA‐seq approaches have revolutionized the detection of cell populations distinguished by their gene expression profiles.⁽ ¹⁴ , ¹⁵ ⁾ The resolution at which rare cell populations can be identified by scRNA‐seq is unmatched compared with bulk‐cell RNA‐seq approaches, leading to the molecular dissection and characterization of cellular heterogeneity within individual tumors,⁽ ¹⁶ , ¹⁷ ⁾ organs,⁽ ¹⁸ ⁾ and entire organisms.⁽ ¹⁹ , ²⁰ ⁾ In addition to detecting distinct cellular populations, scRNA‐seq with time‐series trajectory analysis allows elucidation of molecular transitions and cellular hierarchies throughout somatic cell reprogramming or differentiation.⁽ ²¹ , ²² , ²³ , ²⁴ , ²⁵ ⁾ For example, using human primary myoblasts analyzed by scRNA‐seq throughout differentiation, trajectory analysis identified temporal changes in gene expression and ordered the cells along a “pseudotime” trajectory from undifferentiated to differentiated cell states. Potential also exists for scRNA‐seq to characterize congenital disease pathogenesis, including somatic diseases such as fracture pseudarthroses that may be caused by disruption of osteolineage cell differentiation. To determine whether somatic loss of NF1 in human BSCs results in defects in osteoblast lineage specification, we performed osteogenic time‐series scRNA‐seq analyses using patient‐matched fracture‐ and control bone (ie, iliac crest)‐derived cultured primary cells. Results from our study provide novel insights into the molecular pathogenesis of NF1‐associated pseudarthrosis and demonstrate the feasibility of this approach to investigate somatic skeletal diseases using patient‐derived specimens.

Materials and Methods

Human subjects and cell culture

All subjects provided written informed consent approved by the UT Southwestern Medical Center Institutional Review Board (092011‐034 and 042018‐090). As part of the Scottish Rite for Children Biorepository, excess tissue resected during surgery performed as standard‐of‐care was provided from subjects diagnosed with pseudarthrosis and NF1. Patient‐derived samples included in this study include both sexes and multiple races and ethnicities.

As previously described,⁽ ⁹ ⁾ resected tissue was digested overnight with collagenase at 37°C. The undigested tissues were removed by centrifugation and cells cultured in minimum essential medium‐alpha (MEM‐alpha, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS, Sigma, St. Louis, MO, USA) and 1% penicillin/streptomycin (Thermo Fisher Scientific) for all experiments. The cultured cells were grown to confluence and acquired a fibroblast‐like spindle‐shaped morphology. Cells were washed with phosphate‐buffered saline (Sigma) and passaged using trypsin/EDTA solution (Lonza, Basel, Switzerland). All experiments were performed with early‐passage (<P4) cells.

For osteogenic differentiation of human cells, 5 × 10⁴ cells were cultured in MEM‐alpha media supplemented with 100 μg/ml l‐ascorbic acid 2‐phosphate (Sigma), 10 mM β‐glycerophosphate (Sigma), and 100 nM dexamethasone (Sigma) for either 9 days (scRNA‐seq) or 14 days (single‐cell and bulk‐cell qPCR). Osteogenic media was refreshed every 3 days. Human cells were harvested using RLT plus buffer (Qiagen, Valencia, CA, USA) with beta‐mercaptoethanol (Sigma). Cells were passed through Qiashredder columns and gDNA eliminator columns (Qiagen) before RNA extraction using the RNeasy Plus Mini Kit (Qiagen) following manufacturer's recommendations. The samples were quantified using a Nanodrop1000 spectrophotometer (Thermo Fisher Scientific).

For MEK inhibitor experiments, cells were treated with 2.5 nM trametinib (Selleck Chemicals, Houston, TX, USA).

Animal procedures and husbandry

All animal procedures were approved by the Institutional Animal Care and Use Committee of UT Southwestern Medical Center (2016‐101455). All mice were maintained on the C57BL/6 background. LepR‐cre⁽ ²⁶ ⁾ mice were provided by Jeffrey Friedman. Prrx1‐cre mice were provided by Cliff Tabin. Nf1 ^flox (Nf1 ^tm1Par) and Nf1 ^+/− (Nf1 ^tm1Tyj) were provided by Simon Conway. Nf1 ^LepR mice were generated by breeding LepR‐cre;Nf1 ^flox/flox mice to Nf1 ^+/− mice. Nf1 ^Prrx1 mice were generated by breeding Prrx1‐cre;Nf1 ^+/flox mice to Nf1 ^flox/flox mice. All mice were housed in a conventional facility with ad libitum chow diet and a standard 12‐hour light/dark cycle.

Mouse cell isolation and differentiation

Adult mouse BSCs were flushed from the tibias and femurs of 4‐month‐old mice, expanded in culture, and 1 × 10⁵ cells seeded for all experiments. Cells were cultured in MEM‐alpha (Thermo Fisher Scientific) supplemented with 10% FBS and 1% penicillin/streptomycin (Thermo Fisher Scientific). For osteogenic differentiation, mouse cell cultures were supplemented with 100 μg/mL l‐ascorbic acid 2‐phosphate (Sigma) and 5 mM β‐glycerophosphate (Sigma) for 14 days. Media was refreshed every 3 days. RNA was extracted from cultured cells using TRIzol (Thermo Fisher). cDNA synthesis was performed using the High Capacity RNA to cDNA Kit (Thermo Fisher Scientific). Gene expression of undifferentiated and differentiated samples was evaluated using SYBR Green PCR Master Mix (Thermo Fisher Scientific). Primers sequences are listed in Supplemental Table S1. Target specificity was evaluated by melting curve analysis.

Osteolectin purification

Osteolectin was purified from stable cell lines expressing recombinant 1X FLAG‐tagged Clec11a cDNA, and osteogenic activity of recombinant protein was confirmed using bone marrow stromal cells (BMSCs) cultured from Clec11a‐deficient mice, both as previously described.⁽ ²⁷ ⁾ For osteogenic differentiation, cells were treated with osteogenic supplement and 20 ng/mL recombinant osteolectin.

Single‐cell RNA‐sequencing and analysis

Human primary BSCs from one patient were cultured as described above and harvested before osteogenic differentiation (day 0) and after 3, 6, or 9 days of differentiation. For all experiments, control cells were simultaneously cultured in parallel and differentiated to ensure potency of osteogenic supplement. For each time point, up to 96 cells were isolated and cDNA generated using the Fluidigm C1 instrument and SMARTer Ultra Low RNA Kit (Clontech, Mountain View, CA, USA), respectively. Sequencing libraries were generated using the Nextera XT Library Preparation Kit (Illumina, San Diego, CA, USA; ref #15032354). Single‐cell sequencing libraries of up to 48 cells from each time point were pooled and sequenced using the Illumina HiSeq 2500 generating an average 4.8 M (control) and 5.7 M (fracture) paired‐end 100 bp reads per cell. scRNA‐seq of control and fracture‐derived primary cells were performed independently.

scRNA‐seq analyses were performed using Partek Flow. Sequence reads were mapped to the human reference genome (GRCh38) using HiSat2. Cells with less than 3000 genes detected were excluded, resulting in removal of 0 and 4 cells in the control and fracture experiments, respectively. Gene expression was quantified relative to RefSeq genes and normalized (log₂(counts per million +1)) before analysis. For trajectory analysis, differentially expressed genes (FDRp < 0.01) were detected between the first and last time points using the Partek Gene Specific Analysis (GSA) tool. For fracture cells, differentially expressed genes (FDRp < 0.01) were detected between the first and last time points utilizing only EREG ^LOW cells using GSA. Trajectory analysis was performed using Monocle (v3),⁽ ²⁰ ⁾ and pseudotime was calculated using a root state defined as the trajectory terminus with undifferentiated (day 0) cells. Genes differentially expressed throughout pseudotime, between cell states, or between EREG ^HIGH and EREG ^LOW cells were detected by ANOVA.

Gene set enrichment analysis was performed using the GSEA software⁽ ²⁸ , ²⁹ ⁾ (v4.1.0, Broad Institute, Cambridge, MA, USA). As recommended, significant associations were defined with FDRq < 0.10. Heat maps were generated using core genes identified by GSEA software as those contributing most to the enrichment results.

Single‐cell specific target amplification (STA)

The human fracture cells were cultured and osteogenic differentiation was performed for 9 and 14 days as described above. Cells were harvested using Accutase (Stem Cell Technologies) before and after osteogenesis. Single cells were captured and cDNA synthesized and preamplified using the Fluidigm (Standard BioTools, South San Francisco, CA, USA) C1 instrument and Delta Gene Assays (Fluidigm). Single‐cell samples were genotyped for the patient's somatic mutation, and gene expression analysis was performed using preamplified cDNA samples, including RUNX2 and GAPDH control.

hTERT‐MSC expression analysis

Raw data was downloaded from NCBI GEO (GSE113253) and analyzed using Partek Flow software. Sequence reads were mapped to the human reference genome (hg19) and gene expression quantified relative to RefSeq transcripts. Expression data was normalized (counts per million + 0.0001). One sample (3d adipogenesis, replicate 3) with poor mapping results was excluded. After batch correction for replicate number, genes with low expression across all samples were excluded, and one outlier sample identified by principal component analysis was removed (14d osteogenesis, replicate 3). Differential gene expression analysis was performed using ANOVA.

Western blot analysis

Protein was extracted using RIPA buffer with protease inhibitor cocktail (Roche, Mannheim, Germany) and concentration determined using the standard BCA assay (Thermo Fisher Scientific). Antibody detection was performed using goat anti‐ALPL (1:100) (R&D Systems, Minneapolis, MN, USA) and rabbit anti‐ACTB (1:5000) (Cell Signaling Technologies, Danvers, MA, USA) in 5% nonfat dry milk in TBST buffer overnight at 4°C. After washing, donkey anti‐goat (1:2000) and goat anti‐rabbit (1:5000) IRDye secondary antibodies (LI‐COR Biosciences, Lincoln, NE, USA) were incubated in 5% nonfat dry milk in TBST buffer for 1 hour at room temperature in dark. After washing, images were acquired using the Odyssey CLx system (LI‐COR Biosciences).

Skeletal evaluations

Tibias were harvested from 4 month‐old control and Nf1 ^LepR mice. Bone lengths were measured using calipers. For micro‐CT, the proximal tibias were imaged using a Skyscan 1072 X‐ray microtomograph (Bruker microCT, Kontich, Belgium, software v1.5) set at 50 kV/200 μA. A scout view of each bone was taken and the sample height was adjusted to ensure the bone was within the field of view. The tibia images were obtained at 8‐μm resolution with a rotation step of 0.4° between each image. Solid three‐dimensional models were reconstructed using NRecon (Skyscan, v1.7.4.6), and the trabecular parameters were measured using methods recommended by Skyscan. Regions of interest were analyzed using CTan software (v1.20.3.0). The trabecular parameters were calculated on 200 slices of trabecular bone from a region just below the growth plate with a threshold of 45 to 255, voltage of 50 kV, current of 201 μA, exposure time of 650 ms, 0.5 mm Al filter, average frame of 6, and random movement of 10. Measurements were calculated using the American Society for Bone and Mineral Research nomenclature. Likewise, the cortical parameters were calculated on 100 slices from a region just below the growth plate with a threshold of 102 to 255.

Statistical analysis

Except for time‐series scRNAseq, all analyses of human BSCs included samples from multiple independent and unrelated patients as indicated. Single‐cell STA qPCR analyses were performed using nonparametric Wilcoxon tests. When comparing bulk‐cell patient‐matched control‐ and fracture‐derived samples, significant differences were determined using paired t tests. For analyses of mouse BMSCs, results represent independent experiments performed with different mice at different times. Sample sizes were not predetermined, and mice were randomly assigned to experiments. To assess differences between multiple conditions, such as different groups (ie, genotypes) and treatments (ie, differentiation), we performed two‐way ANOVA with Sidak's multiple comparison. Before ANOVA, normality was assessed using the Shapiro–Wilk test and variance assessed using F test in R. Differences between groups of mice were determined by two‐sided t tests. For Supplemental Fig. S8A , C , one‐sided paired t tests were performed for each genotype after normalization. Most data were normalized using a log transformation. Statistical tests were performed using GraphPad Prism (GraphPad, La Jolla, CA, USA).

Results

Osteogenic trajectory analysis of patient iliac crest‐derived primary cells

We cultured patient iliac crest‐derived BSCs, which do not harbor somatic NF1 gene mutations,⁽ ⁹ ⁾ and performed scRNA‐seq analysis of cells collected before osteogenic differentiation (day 0) and after 3 or 9 days of osteogenesis (Fig. 1A ). Similar to principal component analysis, t‐distributed stochastic neighbor embedding (t‐SNE) plots enable detection of distinct cell types based on single‐cell gene expression profiles. t‐SNE analysis demonstrated cells sampled at all time points were homogenous and without contaminating cell populations (Supplemental Fig. S1 A). Trajectory analysis identified a pseudotime trajectory that correlated with the time at which cells were sampled (Fig. 1B ; Supplemental Fig. S1 B, C). To evaluate whether global changes in the expression of genes relevant to specific biologic processes were associated with cell pseudotime, we performed GSEA. GSEA identified molecular pathways with increased and decreased expression associated with osteogenic pseudotime (Supplemental Tables S2 and S3), including increased expression of genes implicated in osteoblast differentiation (Fig. 1C ; Supplemental Fig. S1D).

Fig. 1 — Single‐cell osteogenic time‐series expression analysis. (A) Schematic of time‐series scRNA‐seq analysis of iliac crest bone stromal cells (BSCs) from a neurofibromatosis type 1 (NF1) patient. BSCs were analyzed before differentiation and after 3 days and 9 days of osteogenic differentiation. (B) Osteogenic pseudotime trajectory analysis of iliac crest‐derived cells. (C) Gene set enrichment analysis (GSEA) identified significant enrichment of genes involved in osteoblast differentiation with increasing expression through pseudotime. (D) Hierarchical clustering of single cells using differentially expressed genes (DEGs) in hTERT‐MSCs after 7 days of osteogenic differentiation. (E) Relative single‐cell expression of genes associated with an osteoblast progenitor cell population after 9 days of differentiation. Cell states correspond to Supplemental Fig. S1 C.

To validate this molecular osteogenic response, we compared scRNA‐seq results of iliac crest‐derived primary cells with those from bulk RNA‐seq analysis of hTERT‐MSCs after 7 days of osteogenic differentiation (Supplemental Fig. S2A, B ).⁽ ¹ ⁾ Using differentially expressed genes (DEGs) identified after 7 days of osteogenic differentiation of hTERT‐MSCs, hierarchical clustering ordered primary patient‐derived cells in a manner consistent with scRNA‐seq trajectory pseudotime (Fig. 1D ). Second, we reciprocally evaluated iliac crest pseudotime‐associated DEGs in undifferentiated and differentiated hTERT‐MSCs. scRNA‐seq pseudotime‐associated DEGs clustered hTERT‐MSCs throughout osteogenesis (Supplemental Fig. S2C ).

Despite a molecular response throughout pseudotime consistent with osteogenic differentiation, the limited time for differentiation (9 days) suggested terminal cells were not fully differentiated osteoblasts but, possibly, represent an intermediate state.⁽ ³⁰ ⁾ We compared our results to cell type‐specific expression biomarkers defined by a publicly available scRNA‐seq data set of mouse FACS‐isolated bone marrow cells.⁽ ² ⁾ Because this data set includes numerous osteolineage cell types representative of intermediate and terminally differentiated cell types in vivo, each with specified expression biomarkers, we investigated the expression of each cell type‐specific biomarker in our scRNA‐seq data set to determine if our differentiated cells represent an intermediate or terminal cell state. Evaluation of these cell type‐specific biomarkers showed that primary cells at the terminus of scRNA‐seq pseudotime are OPCs, maintaining high expression of the stromal C‐X‐C motif chemokine ligand 12 (CXCL12), reduced expression of collagen 1 (COL1A1), and increased expression of periostin (POSTN) (Fig. 1E ). Taken together, time‐series scRNA‐seq analysis detected an osteogenic molecular response of primary patient iliac crest‐derived BSCs to an osteoblast progenitor cell population, consistent with the in vitro osteogenic differentiation potential demonstrated for bulk iliac crest‐derived primary cells (Fig. S1 E).

Trajectory analysis of fracture‐derived NF1 ^+/− primary cells

Prior genetic studies⁽ ⁸ , ⁹ , ³¹ ⁾ of isolated skeletal manifestations in patients with NF1 established a model wherein BSCs cultured from patient fractures are mosaic and consist of both NF1 ^+/− and somatically mutated NF1 ^−/− cells (Fig. 2A ). Using bulk RNA‐seq of mosaic patient fracture‐derived primary cells compared with patient‐matched control cells, we previously reported increased expression of EREG, encoding epiregulin, in fracture‐derived cells.⁽ ⁹ ⁾ We previously reported scRNA‐seq results of clonally isolated and expanded NF1 ^−/− primary cells compared with patient‐matched bulk control bone‐derived cells.⁽ ¹³ ⁾ Clonal scRNA‐seq similarly detected significantly increased EREG expression, further implicating this gene as an expression biomarker of somatically mutated fracture‐derived primary cells. Because of the vast genetic heterogeneity of somatic NF1 gene mutations between patients, it is not feasible to perform single‐cell genotyping without knowing a priori each patient's pseudarthrosis‐associated somatic gene mutation. Alternatively, we utilized relative EREG expression to distinguish NF1 ^−/− somatic cells. To demonstrate the reproducibility of this expression biomarker between patients, we performed qPCR analysis of patient‐matched control (iliac crest)‐ and fracture‐derived BSCs from 6 patients. EREG expression was significantly increased in patient‐matched fracture‐derived BSCs (Fig. 2B ). We then performed time‐series scRNA‐seq analysis using primary patient fracture‐derived cells and sought those cells with elevated EREG expression (Fig. 2C ). A minority cell population with high EREG expression (EREG ^HIGH cells) was evident at all time points, consistent with a somatic NF1 ^−/− cell population (Fig. 2D ; Supplemental Table S4). The majority of fracture‐derived cells at all time points showed low EREG expression (EREG ^LOW cells), similar to patient‐matched iliac crest‐derived NF1 ^+/− cells. Furthermore, the transcriptomes of all cells, including both EREG ^HIGH and EREG ^LOW, were not sufficiently different to be identified as distinct cell clusters by t‐SNE analysis, again confirming the homogeneity in cultured primary cells (Supplemental Fig. S3A, B ).

We tested the molecular response of EREG ^LOW fracture‐derived cells to osteogenic differentiation by pseudotime analysis (Fig. 2E ). Trajectory analysis identified multiple computational states and a pseudotime‐associated OPC expression profile (Supplemental Fig. S3C, D ), similar to patient‐matched iliac crest‐derived cells (Fig. 1). We detected DEGs associated with osteogenic pseudotime of EREG ^LOW fracture‐derived cells (Fig. S3E ). We then tested whether these EREG ^LOW pseudotime‐associated genes could successfully cluster hTERT‐MSCs undergoing osteogenic differentiation, potentially suggesting that the molecular changes associated with pseudotime reflected inherent regulation of osteogenesis. Consistent with this hypothesis, these DEGs successfully distinguished hTERT‐MSCs undergoing osteogenic or adipogenic differentiation (Supplemental Fig. S3F ). GSEA analysis also identified a significant association between the increased expression of genes promoting osteoblast differentiation and EREG ^LOW osteogenic pseudotime (Fig. 2F, G ). Finally, we performed GSEA analysis comparing undifferentiated fracture‐derived EREG ^LOW BSCs with differentiated EREG ^LOW OPCs. Increased expression of multiple gene sets involved in processes required for skeletal development and fracture healing were significantly associated with fracture‐derived EREG ^LOW OPCs, consistent with differentiation to an intermediate osteogenic cell state (Supplemental Fig. S4). These results are consistent with a pro‐osteogenic response of fracture‐derived NF1 ^+/− BSCs co‐cultured with somatic NF1 ^−/− cells and further support the molecular dissection of osteogenic differentiation of patient‐derived primary cells by time‐series scRNA‐seq analyses.

Regulation of osteogenesis in somatic fracture‐derived EREG ^HIGH primary cells

We then used trajectory analysis to investigate the molecular response of EREG ^HIGH fracture‐derived cells during osteogenic differentiation. Despite their reduced number, somatic EREG ^HIGH cells ordered along a linear trajectory (Fig. 3A ), albeit with greater variation compared with trajectory clustering of EREG ^LOW cells (Fig. 2E ). However, in contrast to EREG ^LOW cells, EREG ^HIGH pseudotime was associated with increased expression of genes involved in the negative regulation of osteoblast differentiation (Fig. 3B ) and was not associated (FDRq = 0.15) with increased expression of genes involved in the positive regulation of osteoblast differentiation. To confirm these results, we also performed trajectory analysis combining both EREG ^HIGH and EREG ^LOW cell populations. Both cell populations clustered along a single trajectory (Supplemental Fig. S5A ). Using GSEA analysis of each cell population, we detected differences in the positive enrichment of osteolineage pathways between EREG ^LOW and EREG ^HIGH cell pseudotime (Supplemental Fig. S5B ). Taken together, these results suggest differences in the regulation of osteolineage genes between EREG ^LOW and EREG ^HIGH fracture‐derived cells during osteogenesis.

Fig. 3 — Osteogenic differentiation of neurofibromatosis type 1 (NF1) *EREG* ^HIGH fracture‐derived cells from a NF1 patient. (A) Osteogenic pseudotime trajectory analysis of fracture‐derived *EREG* ^HIGH cells. (B) Gene set enrichment analysis (GSEA) identified significant enrichment of genes involved in the negative regulation of osteoblast differentiation with increasing expression through *EREG* ^HIGH pseudotime. (C) qPCR analysis of *RUNX2* expression in single cells genotyped for somatic *NF1* gene mutations after 9 days and 14 days of osteogenic differentiation. Significant differences were determined by nonparametric Wilcoxon test. Cell numbers per patient and treatment are provided in Supplemental Table S5. (D) Expression of *RUNX2* (left) and *SPP1* (right) in bulk fracture‐derived primary cells from 6 patients before (undifferentiated) and after 14 days (osteogenesis) of osteogenic differentiation. Statistically significant differences were determined using one‐sided paired t test. (E) Expression of *ALPL* in bulk patient‐matched control and fracture‐derived bone stromal cells (BSCs) from 6 patients. Statistically significant differences were determined using two‐sided paired t test. (F) Western blot quantification of ALPL protein in cell lysates from undifferentiated patient‐matched control bone‐ and fracture‐derived bulk BSCs from 4 patients. Statistically significant differences were determined using one‐sided paired t test. (G) Single‐cell expression of *ALPL* in co‐cultured *EREG* ^LOW and *EREG* ^HIGH fracture‐derived cells. Statistically significant differences were determined using ANOVA.

To investigate further the osteogenic potential of NF1 ^−/− fracture‐derived BSCs, we independently differentiated fracture‐derived BSCs from 3 patients for whom the germline and somatic NF1 gene mutations were previously identified.⁽ ⁹ ⁾ Cells were harvested before differentiation and after 9 or 14 days of osteogenesis. The relative expression of the osteogenic marker gene RUNX2 was tested in single cells that were also genotyped for their patient‐specific somatic mutation (Supplemental Table S5). Both fracture‐derived NF1 ^+/− and NF1 ^−/− cells showed significantly increased expression of RUNX2 after both 9 days and 14 days of differentiation (Fig. 3C ). These results were consistent with analysis of bulk fracture‐derived primary cells from multiple patients after 14 days of osteogenic differentiation (Fig. 3D ).

Through the course of differentiation experiments, we observed significantly increased expression of ALPL, the gene encoding alkaline phosphatase, and increased ALPL protein in undifferentiated bulk fracture‐derived BSCs compared with patient‐matched iliac crest‐derived BSCs (Fig. 3E, F ). Increased ALPL expression in EREG ^HIGH fracture‐derived cells was also evident from scRNA‐seq analysis (Fig. 3G ). Increased ALPL expression may be secondary to upregulation of genes in the extracellular pyrophosphate pathway previously reported in mice lacking Nf1 in skeletal cartilage and bone (using Col2a1‐cre).⁽ ³² ⁾ However, we found that expression of the inorganic pyrophosphate transporter (ANKH) was only slightly elevated in fracture‐derived BSCs with no significant difference in expression of the ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) (Supplemental Fig. S5C, D ), as we reported previously.⁽ ³² ⁾ Rather, constitutively increased ALPL expression was evident at all time points (Supplemental Fig. S5E ). Furthermore, increased expression of ALPL in fracture‐derived primary cells was rescued after treatment with the MEK inhibitor trametinib (Supplemental Fig. S5F ). Our results thereby demonstrate that increased ALPL expression is due to altered MEK signaling associated with loss of NF1 in primary fracture‐derived BSCs.

Constitutive somatic dysregulation in EREG ^HIGH fracture‐derived cells

Neurofibromin is a RAS GTPase, and somatic NF1 deficiency leads to activation of the RAS–MEK–ERK signaling pathway.⁽ ⁶ ⁾ Therefore, we investigated the molecular dysregulation inherent in somatic EREG ^HIGH fracture‐derived cells compared with co‐cultured EREG ^LOW cells by scRNA‐seq (Fig. 4A ; Supplemental Table S6). GSEA analysis identified significantly increased expression in EREG ^HIGH cells of genes implicated in cell cycle regulation, ribosome biosynthesis/protein translation, and mitochondrial processes (Supplemental Fig. S6A ; Supplemental Table S7). ERK signaling has been extensively studied for its role in promoting cell cycle/proliferation,⁽ ³³ , ³⁴ ⁾ protein translation,⁽ ³⁵ ⁾ and in regulating mitochondrial metabolism.⁽ ³⁶ ⁾ Next, we evaluated the relative expression of genes significantly upregulated in NF1 ^−/− Schwann cells cultured from NF1 patient plexiform neurofibromas (pNF) compared with normal human Schwann cells.⁽ ³⁷ ⁾ Using area under the curve (AUC) analysis, we detected significantly higher expression of NF1 ^−/− pNF‐associated genes in EREG ^HIGH fracture‐derived cells compared with co‐cultured EREG ^LOW cells (Supplemental Fig. S6B ). These results are consistent with RAS–MEK–ERK pathway activation in somatic fracture‐derived EREG ^HIGH cells.

Fig. 4 — Molecular basis of fracture healing defects associated with *EREG* ^HIGH cells. (A) MA plot of genes differentially expressed between co‐cultured *EREG* ^HIGH and *EREG* ^LOW fracture‐derived cells. (B) Enrichment map summarizing molecular pathways involved in skeletal development associated with reduced expression in *EREG* ^HIGH cells. Edge thickness represents the number of genes shared between gene sets. Node size represents the number of genes in the gene set. (C) Heat map showing single‐cell expression of core genes in molecular pathways (Supplemental Table S9) with significantly increased expression in mouse fractures. Orthologous genes are repressed in somatic *EREG* ^HIGH (green) compared with co‐cultured *EREG* ^LOW (orange) fracture‐derived cells.

Molecular dysregulation of bone mineralization homeostasis in somatic EREG ^HIGH cells

We hypothesized genes required for osteogenic differentiation were expressed at lower levels in EREG ^HIGH cells compared with EREG ^LOW cells, potentially contributing to altered fracture healing in these patients. GSEA analysis detected significantly reduced expression of genes implicated in skeletal development and in the production of extracellular matrix in EREG ^HIGH cells compared with EREG ^LOW cells (Fig. 4B ; Supplemental Fig. S7; Supplemental Table S8). To investigate this in the context of fracture healing, we compared our GSEA results to published RNA‐seq analysis of murine fracture calluses collected 3, 7, or 14 days after fracture.⁽ ³⁸ ⁾ Multiple molecular pathways required for osteochondral differentiation and formation of extracellular matrix were positively enriched in mouse fracture calluses as early as 3 days post‐fracture (Supplemental Table S9). Expression of these gene sets was constitutively reduced in EREG ^HIGH cells (Fig. 4C ; Supplemental Table S9), suggesting deficiencies in bone mineralization and/or extracellular matrix production persist both in NF1 ^−/− BSCs, OPCs, and, potentially, NF1 ^−/− osteoblasts.

Nf1 regulates Clec11a‐Itga11‐Wnt signaling to maintain the adult skeleton

Conditional deletion of Nf1 in mouse embryonic osteochondroprogenitor cells results in severe skeletal deformities due, at least in part, to impaired chondrogenic and osteogenic differentiation.⁽ ³² , ³⁹ ⁾ To test whether loss of Nf1 is required for osteogenic differentiation of adult BMSCs and maintenance of the adult skeleton, we engineered LepR‐cre;Nf1 ^flox/− mice (herein referred to as Nf1 ^LepR). Unlike conditional models targeting embryonic limb mesenchyme (such as Prrx1‐cre;Nf1 ^flox/flox; herein referred to as Nf1 ^Prrx1), LepR‐cre targets adult (>2 months) BMSCs, sparing the growth plate and removing potential effects from altered embryonic chondrogenesis.⁽ ⁴⁰ ⁾ As expected, cultured BMSCs from adult Nf1 ^LepR mice showed significantly reduced expression of Nf1 and increased expression of Ereg compared with BMSCs from LepR‐cre;Nf1 ^+/flox control mice (Fig. 5A ). At 4 months of age, Nf1 ^LepR mice were grossly indistinguishable from littermate controls, though Nf1 ^LepR mice developed slightly reduced body weights and tibia lengths (Fig. 5B ). However, the same Nf1 ^LepR mice developed significantly reduced trabecular bone volume due to reduced trabecular thickness (Fig. 5C ). Additionally, cortical bone volume was significantly reduced in these 4‐month‐old Nf1 ^LepR mice, with significantly increased cortical porosity and decreased cortical thickness compared with littermate controls (Fig. 5D ). We next tested osteogenic differentiation of adult BMSCs cultured from 4‐month‐old control and Nf1 ^LepR mice. After differentiation, expression of the osteogenic genes Dentin matrix protein 1 (Dmp1) and Bone gamma‐carboxyglutamate protein (Bglap) were significantly increased in control BMSCs; however, osteogenic differentiation of Nf1 ^LepR BMSCs was significantly impaired (Figs. 5E, F ). These results demonstrate that Nf1 is required for osteogenic differentiation of adult BMSCs and for the maintenance of the adult skeleton in mice.

Fig. 5 — Altered skeletal development in adult *Nf1* ^LepR mice. (A) Relative expression of *Nf1* (left) and *Ereg* (right) in cultured adult bone marrow stromal cells (BMSCs) from *LepR*‐cre;*Nf1* ^+/flox (control) and *LepR*‐cre;*Nf1* ^flox/− (*Nf1* ^LepR) mice. n = 7–8 per group. Statistically significant differences were determined using two‐sided t test. (B) Body weight (left) and tibia length (right) of 4‐month‐old control and *Nf1* ^LepR mice. n = 5–8 per group. Statistically significant differences were determined by two‐sided t test. (C, D) Micro‐CT analysis of (C) trabecular and (D) cortical bone parameters in the proximal tibia of 4‐month‐old *Nf1* ^LepR and control mice. n = 8–10 per group. Statistically significant differences were determined by two‐sided t test. (E, F) Expression of (E) *Dmp1* and (F) *Bglap* in adult BMSCs from control and *Nf1* ^LepR mice after 14 days of osteogenic differentiation. n = 5 replicates per group. Statistically significant differences were determined using two‐way ANOVA with Sidak multiple comparison test.

Osteolectin, encoded by the C‐type lectin domain containing 11A (Clec11a) gene, is a secreted pro‐osteogenic protein expressed by LepR⁺ adult mouse BMSCs.⁽ ²⁷ ⁾ Clec11a is necessary for osteogenic differentiation, and mice lacking Clec11a (Clec11a ^−/−), or its receptor Integrin alpha 11 (Itga11 ^−/−), develop reduced trabecular bone volume.⁽ ²⁷ , ⁴¹ ⁾ We tested whether recombinant osteolectin (rOln) improved osteogenic differentiation of Nf1 ^LepR BMSCs. After differentiation with rOln, expression of Dmp1 and Bglap were significantly increased in cells cultured from 4‐month‐old control mice, whereas no osteogenic differentiation was evident in cells from Nf1 ^LepR mice (Fig. 6A, B ). Osteolectin promotes osteogenesis through the Itga11 receptor by activating downstream canonical WNT signaling.⁽ ⁴¹ ⁾ Moreover, loss of Nf1 leads to activation of the MAPK signaling pathway that, in the context of cancer, leads to cell type‐specific activation or repression of WNT signaling.⁽ ⁴² ⁾ Thus, we hypothesized the lack of osteogenic response of Nf1 ^LepR BMSCs results from impaired Wnt signaling. To test this, we evaluated expression of WNT‐responsive genes Axin2 and Lef1 in control and Nf1 ^LepR BMSCs after differentiation with rOln. Expression of both Axin2 and Lef1 was significantly increased in control BMSCs after differentiation with rOln (Fig. 6C, D ). Expression of Axin2 was significantly increased in Nf1 ^LepR BMSCs treated with rOln, whereas no difference in Lef1 expression was detected.

Fig. 6 — Loss of *Nf1* alters activation of Wnt signaling during differentiation. (A–D) Expression of (A) *Dmp1*, (B) *Bglap*, (C) *Axin2*, and (D) *Lef1* in cultured adult bone marrow stromal cells (BMSCs) from control and *Nf1* ^LepR mice after 14 days of osteogenic differentiation with recombinant osteolectin (rOln). n = 5 replicates per group. Statistically significant differences were determined by two‐way ANOVA with Sidak multiple comparison test. (E, F) Expression of (E) *Itga11* and (F) *Clec11a* in cultured adult bone marrow stromal cells (BMSCs) from control and *Nf1* ^LepR mice. n = 5 replicates per group. Statistically significant differences were determined by two‐sided t test. (G, H) Single‐cell expression of (G) *ITGA11* and (H) *CLEC11A* in fracture‐derived *EREG* ^LOW and *EREG* ^HIGH cells. Statistically significant differences were determined by ANOVA.

We next sought to replicate these results using BMSCs from Nf1 ^Prrx1 mice. Compared with control, osteogenic differentiation of BMSCs from 2‐month‐old Nf1 ^Prrx1 mice was significantly impaired (Supplemental Fig. S8A, B ). Consistent with results from Nf1 ^LepR mice, BMSCs from Nf1 ^Prrx1 mice failed to differentiate in response to rOln, with little evidence for Wnt pathway activation (Supplemental Fig. S8C–F ).

Finally, we hypothesized that loss of Nf1 results in dysregulation of the Clec11a‐Itga11‐Wnt signaling pathway, leading to impaired osteogenic differentiation in response to rOln and skeletal disease in these mice. We tested the expression of both Clec11a and Itga11 in BMSCs from 4‐month‐old control and Nf1 ^LepR mice. Although no difference in the expression of Clec11a was observed, expression of Itga11 was significantly reduced in cells from Nf1 ^LepR mice compared with controls (Fig. 6E, F ). To test whether similar differences were evident in somatic patient fracture‐derived cells, we evaluated these genes in EREG ^HIGH and EREG ^LOW fracture‐derived cells. Expression of ITGA11, but not CLEC11A, was also significantly reduced in patient fracture‐derived EREG ^HIGH cells compared with EREG ^LOW cells (Fig. 6G, H ; Supplemental Fig. S9).

Discussion

scRNA‐seq enables the molecular dissection of cell types throughout entire organisms,⁽ ¹⁹ ⁾ of distinct hierarchical cell states throughout differentiation,⁽ ⁴³ , ⁴⁴ ⁾ and of cellular heterogeneity within and between human malignancies.⁽ ⁴⁵ ⁾ However, similar studies to elucidate the molecular pathogenesis of non‐cancer somatic diseases have been limited. Here, we performed time‐series scRNA‐seq of primary BSCs cultured from control bone and fracture pseudarthrosis from a patient with NF1. Our results suggest a shared molecular pathology of NF1 pseudarthroses and pNF tumors, suggesting a potential for targeted therapies, such as MEK inhibitors,⁽ ⁴⁶ , ⁴⁷ ⁾ in the treatment of NF1 pseudarthrosis. Consistent with this, MEK‐dependent activation of the pyrophosphate pathway was identified in both NF1‐associated somatic skeletal and tumor manifestations.⁽ ³² , ⁴⁸ ⁾ Previous studies have implicated defects in osteogenic differentiation contributing to Nf1‐associated skeletal disease in mice.⁽ ³² , ³⁹ , ⁴⁹ ⁾ Our time‐series analysis demonstrates that osteogenic specification of somatic NF1 ^−/− BSCs to OPCs remains intact. Rather, we identified molecular signatures converging on constitutively altered skeletal mineralization in the pathogenesis of somatic NF1 pseudarthrosis. Consistent with this, we show that Nf1 is required in LepR‐expressing BMSCs for the maintenance of the adult mouse skeleton. Finally, we show that reduced ITGA11 expression is conserved in patient fracture‐derived somatic EREG ^HIGH cells compared with EREG ^LOW cells, implicating defects in the CLEC11A‐ITGA11‐WNT signaling pathway in the pathogenesis of somatic NF1 skeletal disease. Taken together, scRNA‐seq analysis of patient‐derived BSCs enabled dissection of molecular pathways associated with somatic skeletal disease in NF1.

The single‐cell approach applied here is broadly applicable to any patient‐derived cell‐based model of disease, ushering a new “disease in a dish” approach to functionally investigate genetic mechanisms of inherited and somatic skeletal disease. Such an approach may be particularly suited to somatic skeletal diseases, where primary cells cultured from patients' pathologic bone are mosaic and consist of small fractions of cells harboring disease‐causing somatic mutations. Single‐cell analyses provide unique opportunities to disambiguate otherwise homogenous somatic cell populations in vitro using expression biomarkers, or through simultaneous single‐cell genotyping, to functionally evaluate molecular dysregulation leading to specification and/or differentiation defects.

Despite these opportunities, challenges related to scRNA‐seq and the study of rare somatic cell populations persist. First, scRNA‐seq fails to capture the broader depth of expression profiling compared with bulk‐cell RNA‐seq approaches. Therefore, it is possible that key genes contributing to disease pathology may be missed by this approach, either because their expression is below the sensitivity of scRNA‐seq methodologies or because their expression is transient. Although sensitivity to detect genes with lower levels of expression will improve with continued development of scRNA‐seq technologies, detecting transient molecular events may require analysis of additional time points with greater numbers of somatic cells. Second, independent validation is necessary to confirm scRNA‐seq discoveries. Here, we highlight the use of independent patient samples and the integration with preclinical mouse models to support results from somatic scRNA‐seq analyses. Finally, translating results from cultured primary cells to patient fracture tissue remains a persistent challenge. However, emerging technologies, such as spatial transcriptomics, may enable analysis of patient samples in a more physiologic context and provide further support of discoveries from in vitro and other preclinical disease models.

Similar approaches may also successfully delineate molecular responses of patient‐derived primary cells to candidate therapies in vitro, which may complement in vivo preclinical studies and support predictions of translational efficacy in human trials.⁽ ⁵⁰ ⁾ Finally, we anticipate single‐cell time‐series trajectory analysis using patient‐derived primary cells will elucidate disease etiologies associated with lineage specification and differentiation of other skeletal cell types (ie, chondrocytes). Using patient‐derived primary cells, this approach will enhance our understanding of skeletal progenitor cell differentiation and molecular mechanisms of skeletal disease.

Conflicts of Interest

The authors declare no competing interests.

Peer Review

The peer review history for this article is available at https://publons.com/publon/10.1002/jbmr.4755.

Supporting information

Supplemental Fig. S1. Molecular analysis of iliac crest‐derived primary cells from a NF1 patient. (A) t‐SNE plot of cultured patient iliac crest‐derived single cells analyzed before differentiation (day 0) and after 3 or 9 days of osteogenic differentiation. (B) Boxplot demonstrating the relationship between trajectory pseudotime and the time point during differentiation at which cells were analyzed. (C) Osteogenic trajectory of iliac crest‐derived primary cells indicating the computational cell state (left) and the relative proportion of cells at each time point for each state (right). (D) Heat map and unsupervised clustering showing single‐cell expression of core genes in the Osteogenic Differentiation gene set. Trajectory pseudotime is shown for each cell. (E) Relative expression of RUNX2, ALPL, and SPP1 in iliac crest‐derived bulk primary cells before differentiation (Undifferentiated) and after 14 days of osteogenic differentiation (Osteogenesis). n = 6 patients. Statistically significant differences were determined by two‐sided paired t test.

Supplemental Fig. S2. Comparison of osteogenic pseudotime to bulk RNAseq. (A) PCA analysis of published⁽ ¹ ⁾ hTERT‐MSC RNA‐seq before differentiation (Undiff; red) and after 3 days (blue), 7 days (green), or 14 days (orange) of osteogenesis (Osteo.). (B) Volcano plot showing genes with significantly (FDRp < 0.05) increased (>2‐fold; red) or decreased (<−2 fold; blue) expression after 7 days of osteogenic differentiation compared with undifferentiated hTERT‐MSCs. (C) Hierarchical clustering and heat map of hTERT‐MSCs using genes differentially expressed throughout osteogenic pseudotime of iliac crest‐derived primary cells.

Supplemental Fig. S3. Differential regulation of osteogenic and adipogenic signatures in EREG ^LOW fracture‐derived cells. (A, B) t‐SNE plot of all fracture‐derived single cells distinguished by their (A) EREG expression status or (B) osteogenic time point at which they were harvested. (C) Trajectory analysis identified multiple computationally defined states throughout pseudotime of NF1 EREG ^LOW fracture‐derived primary cells. (D) Expression of OPC‐associated genes in EREG ^LOW fracture‐derived cells. (E) Hierarchical clustering and heatmap of genes differentially expressed through pseudotime (FDRp < 0.05). (F) Hierarchical clustering and heat map of hTERT‐MSCs before differentiation and throughout osteogenic or adipogenic differentiation using EREG ^LOW pseudotime‐associated genes.

Supplemental Fig. S4. Gene set enrichment analysis of EREG ^LOW fracture‐derived OPCs. (A–D) Gene set enrichment analysis (left) and heat map of core enrichment genes (right) for indicated gene ontologies associated with increased expression in EREG ^LOW fracture‐derived OPCs compared with BSCs.

Supplemental Fig. S5. Altered gene expression associated with somatic patient fracture‐derived primary cells. (A) Trajectory analysis of all fracture‐derived primary cells. EREG ^LOW cells are gray, and somatic EREG ^HIGH cells are indicated by the time point at which they were harvested. (B) Results for related gene sets after pseudotime gene set enrichment analysis (GSEA) of EREG ^LOW (blue) and EREG ^HIGH (red) cells. Pseudotime was calculated from the combined trajectory in (A). (C, D) Relative expression of (C) ANKH and (D) ENPP1 in patient‐matched control bone‐ and fracture‐derived bulk primary cells. n = 6 patients. Statistically significant differences were determined by two‐sided paired t test. (E) Relative single‐cell expression of ALPL before differentiation (day 0) or after 3, 6, or 9 days of osteogenic differentiation. (F) Relative ALPL expression in undifferentiated fracture‐derived primary cells without (Control) or with 2.5 nm trametinib (MEKi). Data represent 2–3 replicates from n = 2 patients. Statistically significant differences were determined by two‐sided paired t test.

Supplemental Fig. S6. Activated gene expression in EREG ^HIGH fracture‐derived cells. (A) Enrichment map summarizing molecular pathways associated with increased expression in EREG ^HIGH cells. Edge thickness represents the number of genes shared between gene sets. Node size reflects the number of genes within the gene set. (B) AUC analysis of EREG ^LOW and EREG ^HIGH fracture‐derived primary cells using genes previously identified with increased expression in NF1 ^−/− Schwann cells.⁽ ³⁷ ⁾ Statistically significant differences were determined by ANOVA.

Supplemental Fig. S7. Developmental gene networks repressed in EREG ^HIGH fracture‐derived cells. (A–C) Enrichment maps summarizing molecular pathways significantly repressed in somatic EREG ^HIGH cells compared with co‐cultured EREG ^LOW cells. Edge thickness represents the number of genes shared between gene sets. Node size reflects the number of genes within the gene set.

Supplemental Fig. S8. Osteogenic response of Nf1 ^Prrx1 BMSCs to recombinant osteolectin. (A, B) Relative expression of (A) Dmp1 and (B) Bglap after 14 days of osteogenic differentiation. Statistically significant differences were determined by one‐sided paired t test (A) or two‐way ANOVA with Sidak multiple comparison test (B). (C–F) Relative expression of (C) Dmp1, (D) Bglap, (E) Axin2, and (F) Lef1 after 14 days of osteogenic differentiation with recombinant osteolectin (rOln). Statistically significant differences were determined by one‐sided paired t test (C) or two‐way ANOVA with Sidak multiple comparison test (D–F).

Supplemental Fig. S9. Reduced ITGA11 expression in fracture‐derived EREG ^HIGH cells. (A–D) Relative single‐cell expression of ITGA11 (A) before differentiation or after (B) 3, (C) 6, or (D) 9 days of osteogenic differentiation. Statistically significant differences were determined by ANOVA.

Click here for additional data file.^{(3.3MB, pdf)}

Supplemental Table S1. Primer Sequences for Expression Analysis

Supplemental Table S2. Gene Set Enrichment Results for Positive Associations With Osteogenic Pseudotime in Iliac Crest Cells

Supplemental Table S3. Gene Set Enrichment Results for Negative Associations With Osteogenic Pseudotime in Iliac Crest Cells

Supplemental Table S4. Proportion of Fracture‐Derived Cells by EREG Expression at Each Osteogenic Time Point

Supplemental Table S5. Number of Single Cells Genotyped and Tested for Osteogenic Differentiation

Supplemental Table S6. Genes Differentially Expressed Between Fracture‐Derived EREG‐High and EREG‐Low Cells

Supplemental Table S7. Gene Set Enrichment Results for Significant (FDRq < 0.10) Positive Association With EREG‐HIGH Cells

Supplemental Table S8. Gene Set Enrichment Results for Significant (FDRq < 0.10) Negative Association With EREG‐HIGH Cells

Supplemental Table S9. Comparison of Gene Sets Significantly Repressed in EREG‐HIGH Cells and Activated in Murine Fracture Calluses

Click here for additional data file.^{(402.7KB, xlsx)}

Acknowledgments

This research was supported by the National Institutes of Health (U54CA196519), the Department of Defense (W81XWH‐18‐1‐0817), the Pediatric Orthopedic Society of North America (POSNA), the Texas Neurofibromatosis Foundation (all to JJR), and the Scottish Rite for Children Research Fund. The authors thank Kristin Denton, Florent Elefteroiu, Dr Bret Evers, Jinghua Gu, Sean Morrison, Dr B Stevens Richards, John Shelton, and Zhiyu Zhao for helpful discussion and/or technical assistance. The authors acknowledge the support of all orthopedic surgeons at Scottish Rite for Children and the resources of the Scottish Rite for Children Biorepository.

Data Availability Statement

hTERT‐MSC RNA‐seq data (GSE113253)⁽ ¹ ⁾ and mouse fracture callus RNA‐seq results (GSE152677)⁽ ³⁸ ⁾ were downloaded from the NCBI Gene Expression Omnibus. Time‐series scRNA‐seq data is available via the NCBI Gene Expression Omnibus (GSE196652).

References

1. Rauch A, Haakonsson AK, Madsen JGS, et al. Osteogenesis depends on commissioning of a network of stem cell transcription factors that act as repressors of adipogenesis. Nat Genet. 2019;51(4):716‐727. [DOI] [PubMed] [Google Scholar]
2. Wolock SL, Krishnan I, Tenen DE, et al. Mapping distinct bone marrow niche populations and their differentiation paths. Cell Rep. 2019;28(2):302‐11 e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
3. Yue R, Zhou BO, Shimada IS, Zhao Z, Morrison SJ. Leptin receptor promotes adipogenesis and reduces osteogenesis by regulating mesenchymal stromal cells in adult bone marrow. Cell Stem Cell. 2016;18(6):782‐796. [DOI] [PubMed] [Google Scholar]
4. de la Croix NJ, Stevens DM, Vignaux G, et al. Combined MEK inhibition and BMP2 treatment promotes osteoblast differentiation and bone healing in Nf1Osx −/− mice. J Bone Miner Res. 2015;30(1):55‐63. [DOI] [PMC free article] [PubMed] [Google Scholar]
5. Julien A, Perrin S, Duchamp de Lageneste O, et al. FGFR3 in periosteal cells drives cartilage‐to‐bone transformation in bone repair. Stem Cell Rep. 2020;15(4):955‐967. [DOI] [PMC free article] [PubMed] [Google Scholar]
6. Gutmann DH, Ferner RE, Listernick RH, Korf BR, Wolters PL, Johnson KJ. Neurofibromatosis type 1. Nat Rev Dis Primers. 2017;3:17004. [DOI] [PubMed] [Google Scholar]
7. Crawford AH, Schorry EK. Neurofibromatosis update. J Pediatr Orthop. 2006;26(3):413‐423. [DOI] [PubMed] [Google Scholar]
8. Stevenson DA, Zhou H, Ashrafi S, et al. Double inactivation of NF1 in tibial pseudarthrosis. Am J Hum Genet. 2006;79(1):143‐148. [DOI] [PMC free article] [PubMed] [Google Scholar]
9. Paria N, Cho TJ, Choi IH, et al. Neurofibromin deficiency‐associated transcriptional dysregulation suggests a novel therapy for tibial pseudoarthrosis in NF1. J Bone Miner Res. 2014;29(12):2636‐2642. [DOI] [PMC free article] [PubMed] [Google Scholar]
10. Lee SM, Choi IH, Lee DY, et al. Is double inactivation of the Nf1 gene responsible for the development of congenital pseudarthrosis of the tibia associated with NF1? J Orthop Res. 2012;30(10):1535‐1540. [DOI] [PubMed] [Google Scholar]
11. Brekelmans C, Hollants S, De Groote C, et al. Neurofibromatosis type 1‐related pseudarthrosis: beyond the pseudarthrosis site. Hum Mutat. 2019;40(10):1760‐1767. [DOI] [PubMed] [Google Scholar]
12. Nahm NJ, Makarewich CA, Rosenwasser KA, Herzenberg JE, McClure PK. Does an osteotomy performed in congenital pseudarthrosis of the tibia heal? J Pediatr Orthop. 2022;42(6):e630–e5. [DOI] [PubMed] [Google Scholar]
13. Tahaei SE, Couasnay G, Ma Y, et al. The reduced osteogenic potential of Nf1‐deficient osteoprogenitors is EGFR‐independent. Bone. 2018;106:103‐111. [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Tang F, Barbacioru C, Wang Y, et al. mRNA‐Seq whole‐transcriptome analysis of a single cell. Nat Methods. 2009;6(5):377‐382. [DOI] [PubMed] [Google Scholar]
15. Shapiro E, Biezuner T, Linnarsson S. Single‐cell sequencing‐based technologies will revolutionize whole‐organism science. Nat Rev Genet. 2013;14(9):618‐630. [DOI] [PubMed] [Google Scholar]
16. Patel AP, Tirosh I, Trombetta JJ, et al. Single‐cell RNA‐seq highlights intratumoral heterogeneity in primary glioblastoma. Science. 2014;344(6190):1396‐1401. [DOI] [PMC free article] [PubMed] [Google Scholar]
17. Puram SV, Tirosh I, Parikh AS, et al. Single‐cell transcriptomic analysis of primary and metastatic tumor ecosystems in head and neck cancer. Cell. 2017;171(7):1611‐24 e24. [DOI] [PMC free article] [PubMed] [Google Scholar]
18. de Soysa TY, Ranade SS, Okawa S, et al. Single‐cell analysis of cardiogenesis reveals basis for organ‐level developmental defects. Nature. 2019;572(7767):120‐124. [DOI] [PMC free article] [PubMed] [Google Scholar]
19. Cao J, Packer JS, Ramani V, et al. Comprehensive single‐cell transcriptional profiling of a multicellular organism. Science. 2017;357(6352):661‐667. [DOI] [PMC free article] [PubMed] [Google Scholar]
20. Cao J, Spielmann M, Qiu X, et al. The single‐cell transcriptional landscape of mammalian organogenesis. Nature. 2019;566(7745):496‐502. [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Wagner DE, Klein AM. Lineage tracing meets single‐cell omics: opportunities and challenges. Nat Rev Genet. 2020;21(7):410‐427. [DOI] [PMC free article] [PubMed] [Google Scholar]
22. Treutlein B, Lee QY, Camp JG, et al. Dissecting direct reprogramming from fibroblast to neuron using single‐cell RNA‐seq. Nature. 2016;534(7607):391‐395. [DOI] [PMC free article] [PubMed] [Google Scholar]
23. Fletcher RB, Das D, Gadye L, et al. Deconstructing olfactory stem cell trajectories at single‐cell resolution. Cell Stem Cell. 2017;20(6):817‐30 e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
24. Liu Z, Wang L, Welch JD, et al. Single‐cell transcriptomics reconstructs fate conversion from fibroblast to cardiomyocyte. Nature. 2017;551(7678):100–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Velasco S, Ibrahim MM, Kakumanu A, et al. A multi‐step transcriptional and chromatin state cascade underlies motor neuron programming from embryonic stem cells. Cell Stem. Cell. 2017;20(2):205‐17 e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
26. DeFalco J, Tomishima M, Liu H, et al. Virus‐assisted mapping of neural inputs to a feeding center in the hypothalamus. Science. 2001;291(5513):2608‐2613. [DOI] [PubMed] [Google Scholar]
27. Yue R, Shen B, Morrison SJ. Clec11a/osteolectin is an osteogenic growth factor that promotes the maintenance of the adult skeleton. Elife. 2016;5:e18782. [DOI] [PMC free article] [PubMed] [Google Scholar]
28. Subramanian A, Tamayo P, Mootha VK, et al. Gene set enrichment analysis: a knowledge‐based approach for interpreting genome‐wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545‐15550. [DOI] [PMC free article] [PubMed] [Google Scholar]
29. Mootha VK, Lindgren CM, Eriksson KF, et al. PGC‐1alpha‐responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet. 2003;34(3):267‐273. [DOI] [PubMed] [Google Scholar]
30. Krause U, Seckinger A, Gregory CA. Assays of osteogenic differentiation by cultured human mesenchymal stem cells. Methods Mol Biol. 2011;698:215‐230. [DOI] [PubMed] [Google Scholar]
31. Margraf RL, VanSant‐Webb C, Mao R, et al. NF1 somatic mutation in dystrophic scoliosis. J Mol Neurosci. 2019;68(1):11‐18. [DOI] [PubMed] [Google Scholar]
32. de la Croix NJ, Makowski AJ, Uppuganti S, et al. Asfotase‐alpha improves bone growth, mineralization and strength in mouse models of neurofibromatosis type‐1. Nat Med. 2014;20(8):904‐910. [DOI] [PMC free article] [PubMed] [Google Scholar]
33. Zhang W, Liu HT. MAPK signal pathways in the regulation of cell proliferation in mammalian cells. Cell Res. 2002;12(1):9‐18. [DOI] [PubMed] [Google Scholar]
34. Lewis TS, Shapiro PS, Ahn NG. Signal transduction through MAP kinase cascades. Adv Cancer Res. 1998;74:49‐139. [DOI] [PubMed] [Google Scholar]
35. Roux PP, Topisirovic I. Regulation of mRNA translation by signaling pathways. Cold Spring Harb Perspect Biol. 2012;4(11):a012252. [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Masgras I, Ciscato F, Brunati AM, et al. Absence of neurofibromin induces an oncogenic metabolic switch via mitochondrial ERK‐mediated phosphorylation of the chaperone TRAP1. Cell Rep. 2017;18(3):659‐672. [DOI] [PubMed] [Google Scholar]
37. Pemov A, Li H, Patidar R, et al. The primacy of NF1 loss as the driver of tumorigenesis in neurofibromatosis type 1‐associated plexiform neurofibromas. Oncogene. 2017;36(22):3168‐3177. [DOI] [PMC free article] [PubMed] [Google Scholar]
38. Coates BA, McKenzie JA, Buettmann EG, et al. Transcriptional profiling of intramembranous and endochondral ossification after fracture in mice. Bone. 2019;127:577‐591. [DOI] [PMC free article] [PubMed] [Google Scholar]
39. Kolanczyk M, Kossler N, Kuhnisch J, et al. Multiple roles for neurofibromin in skeletal development and growth. Hum Mol Genet. 2007;16(8):874‐886. [DOI] [PubMed] [Google Scholar]
40. Zhou BO, Yue R, Murphy MM, Peyer JG, Morrison SJ. Leptin‐receptor‐expressing mesenchymal stromal cells represent the main source of bone formed by adult bone marrow. Cell Stem Cell. 2014;15(2):154‐168. [DOI] [PMC free article] [PubMed] [Google Scholar]
41. Shen B, Vardy K, Hughes P, et al. Integrin alpha11 is an osteolectin receptor and is required for the maintenance of adult skeletal bone mass. Elife. 2019;8:e42274. [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Guardavaccaro D, Clevers H. Wnt/beta‐catenin and MAPK signaling: allies and enemies in different battlefields. Sci Signal. 2012;5(219):pe15. [DOI] [PubMed] [Google Scholar]
43. Nestorowa S, Hamey FK, Pijuan Sala B, et al. A single‐cell resolution map of mouse hematopoietic stem and progenitor cell differentiation. Blood. 2016;128(8):e20‐e31. [DOI] [PMC free article] [PubMed] [Google Scholar]
44. Buenrostro JD, Corces MR, Lareau CA, et al. Integrated single‐cell analysis maps the continuous regulatory landscape of human hematopoietic differentiation. Cell. 2018;173(6):1535‐48 e16. [DOI] [PMC free article] [PubMed] [Google Scholar]
45. Qian J, Olbrecht S, Boeckx B, et al. A pan‐cancer blueprint of the heterogeneous tumor microenvironment revealed by single‐cell profiling. Cell Res. 2020;30(9):745‐762. [DOI] [PMC free article] [PubMed] [Google Scholar]
46. Dombi E, Baldwin A, Marcus LJ, et al. Activity of selumetinib in neurofibromatosis type 1‐related plexiform neurofibromas. N Engl J Med. 2016;375(26):2550‐2560. [DOI] [PMC free article] [PubMed] [Google Scholar]
47. Gross AM, Wolters PL, Dombi E, et al. Selumetinib in children with inoperable plexiform neurofibromas. N Engl J Med. 2020;382(15):1430‐1442. [DOI] [PMC free article] [PubMed] [Google Scholar]
48. Ma Y, Gross AM, Dombi E, et al. A molecular basis for neurofibroma‐associated skeletal manifestations in NF1. Genet Med. 2020;22(11):1786–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Wu X, Chen S, He Y, et al. The haploinsufficient hematopoietic microenvironment is critical to the pathological fracture repair in murine models of neurofibromatosis type 1. PLoS One. 2011;6(9):e24917. [DOI] [PMC free article] [PubMed] [Google Scholar]
50. Hackam DG, Redelmeier DA. Translation of research evidence from animals to humans. JAMA. 2006;296(14):1731‐1732. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials